Introduction

Data Extraction

Airbnb daily data is very valuable. An investor is eager to use this data to make key decisions about the best real estate option available to generate benefit. In this project, we will use scrapped data from 6 countries aggregated over a period of time (For each city of these countries we kept the 3 latest dates collected):

• France (from 2020-06-11 to 2020-09-19)
• Netherlands (from 2020-07-09 to 2020-09-09)
• Spain (from 2020-04-30 to 2020-09-19)
• Italy (from 2020-04-28 to 2020-09-25)
• Germany (from 2020-04-25 to 2020-08-30)
• Belgium (from 2020-04-19 to 2020-06-22)

These datasets can be downloaded on this website http://insideairbnb.com/get-the-data.html. A csv document is available in data\all_data_urls.csv where all the scrapped urls are available and ready to download.

Prepocessing overview

Because these datasets are huge, we made some processing to focus on important information and at the same time use a reasonable amount of data (fit computation and time limitations). We’ve been through several steps:

################### Code From utils/tools.R ##################################
urls <- read.csv(file.path("./data/all_data_urls.csv")) # Step 1 
df <- extract_all_meta(urls) # Step 2 
lastest_dates <- 3 # Step 3
countries <- c("france", "spain", "the-netherlands", "germany", "belgium","italy") # Step 4
download_data(df, countries, lastest_dates) # Step 5

listings <- load_global_listings() # Step 6

We reduce the data size from several Gb to only a hundred of Mb. We are now ready to play with it!

Prepocessing Steps

Starting with raw data, we’ve been through several steps:

[Step 1] Load csv data with urls and meta provided (read.csv)
[Step 2] Extract “country”, “region”, “city”, “date” and “url” from the csv in a dataframe (extract_all_meta)
[Step 3] Specify the number “n” of latest scrapping dates you are looking for.
[Step 4] Select a list of 6 countries, on which you want to work on.
[Step 5] Go through this dataframe, line by line and do the following steps (download_data and prepare_data) :

• For each country of the list:
  • Filter the dataframe (from extract_all_meta) to get the country
  • Order then by date and keep the n most recent scrapped dates
  • For each remaining rows (one city per line) download both calendar and listings, add the metadata to both, send it to prepare_data.
    • Prepare_data is pretty much the same script as the teacher’s (with more columns kept)
    • Select interesting columns (listed in previous slides)
    • Cleaning calendar dataframe by:
    • Arranging by id and date
    • Adding day number
    • Changing available column to binary
    • Cleaning price column and transform to numeric
    • Calculating estimated availability, price, revenue for the next 30 for each listing_id
    • Finally concatenateing listings and agenda together into one dataframe
    • Store the dataframe in a csv file called /data/countries/listings_CITY_NAME_date.csv.

[Step 5 - Remarque] This step results in a big csv file for every cities of the countries listed.
We could have written the csv’s into files, but since this step takes more than 10 minutes, we preferred to keep them in memory.

[Step 6] Get Final preprocessed dataset by merging all the cities csv into a single data frame (load_global_listings).
This step is performed when the server starts and takes around 20 seconds.

Data Overview

Here is the shape of our dataset:
# Publications :1321825
# Features :21

Feature names are:

## - id
## - country
## - region
## - city
## - date
## - neighbourhood_cleansed
## - latitude
## - longitude
## - property_type
## - room_type
## - accommodates
## - bedrooms
## - beds
## - price
## - minimum_nights
## - maximum_nights
## - review_scores_rating
## - availability_30
## - price_30
## - revenue_30
## - latitudelongitude

Shiny App: Tab1 - Analysis by comparing several cities

Tab 1 - Analysis by comparing several cities

Shiny App: Tab2 - Analysis only one city

Tab 2 - Analysis only one city

Shiny App: Structure and code overview

Libraries

We used several libraries (webapp, graphical, data manipulation) to build this project:
shiny, googleVis, ggplot2, dplyr, data.table, stringr and glue

UI

################### Code From shinyApp/ui.R ##################################
# IT IS SPEUDO CODE !!!
fluidPage
  tabsetPanel
    tabPanel # Analysis 1 Tab
      sidebarLayout
        sidebarPanel # Tool Bar
          Checkbox, selectInput, uiOutput, ...
        mainPanel # Plots
          htmlOutput, plotOutput ...
    tabPanel # Analysis 2 Tab
      sidebarLayout
        sidebarPanel # Tool Bar
          Checkbox, selectInput, uiOutput, ...
        mainPanel # Plots
          htmlOutput, plotOutput ...

Server

################### Code From shinyApp/server.R ##################################
# IT IS SPEUDO CODE !!!
listings <- load_global_listings() # Download data
# Server
server
  # Tab 1 variables
  reactive # Reactive DataFrame (filter by country / cities / features)
  renderUI  # ui send from server to uiOutput (checkbox, selectInput, dateSlider)
  renderGvis, renderPlot # Plots send to ui send from server to htmlOutput,plotOutput (histogram,...)
  # Tab 2 variables
  reactive # Reactive DataFrame (filter one city)
  renderUI  # ui send from server to uiOutput (checkbox, selectInput, dateSlider)
  renderGvis, renderPlot # Plots  send to ui from server to htmlOutput,plotOutput (map,...)

App Usage: Analysis 1 Comparing Cities

Each tab is split into two vertical parts: Tool Bar and Plots

Tool Bar

You can:

• Select a country (France, Netherlands, Spain, Italy, Germany or Belgium) (will be kept in tab2).
• Select cities to compare (initially none are selected, thus no plots are displayed).
• Select a range of date (might produce unexpected results since we only keep 3 scrapping dates).
• Select a numerical feature (1) (Revenue, Availability, Price, Minimum Night, Maximum Nights, Review Scores Rating).
• Select a categorical feature (2) (Room type, Number of bedrooms).

Plots

• First plot is a histogram displaying the average of feature 1 for each selected city
• Second plot is a density plot of feature 1 for each selected city
• Last plot is a boxplot of feature 1 and feature 2 for each selected city

App Usage: Analysis 2 Deep Dive in one City

Tool Bar

You can:

• Select only one city (from the country selected in tab1).
• Select a range of date (can be different from tab1).
• Select a numerical feature (1) (Revenue, Availability, Price, Minimum Night, Maximum Nights, Review Scores Rating).
• Select a categorical feature (2) (Room type, Number of bedrooms, Neighbourhood).
• Select map filters (Minimum value of feature 1, Maximun value of feature 1 and the Number of geopoints to display).

Plots

• First two plots are histograms displaying:
• The Proportion of feature 2.
• The Average value of feature 1 for each feature 2 categories.
• Second plot is a Density plot and average line of feature 1 for each feature 2 categories.
• Last plot is a Google Map specifying the feature 1 value for each geopoints based on the map filters selected.

Let’s try this in our Shiny App

library(shiny) # You might need to install more packages (ggplot, gvis, ...)

setwd("~/YOUR_PATH/Airbnb-Analysis-ShinyApp")
runApp(appDir = "shinyApp")

#or 
setwd("~/YOUR_PATH/Airbnb-Analysis-ShinyApp")
runGitHub("Airbnb-Analysis-ShinyApp", "MarvinMartin24", subdir = "shinyApp")